WOW - what a tale - I was having trouble breathing !!


Not sure what to do with these old war stories,
the original intent of my web site was hardware.

Any yours was more a very good real life learning experience
than just the normal techie tale  :-))

I am appending a little of my interaction with LGP-30s.

I'm forwarding our tales to a computer tale list
and I'm linking to it from my LGP-30 page.

Thanks much
   Ed Thelen


---------------------------------------------------


----- Original Message ----- 
From: "Bill Cole 3" 
To: 
Cc: "Synod" 
Sent: Friday, February 25, 2005 11:45 AM
Subject: An LGP-30 war story from 1959-60


> Hiya, Ed  --
> 
> Appreciate your treatment of computing stone ages, ref.
> http://ed-thelen.org/comp-hist/lgp-30.html
> 
> Would anyone be interested in an anecdotal story of making LGP-30 do 
> something useful in 6-8 weeks from a standing start  --  something 
> absolutely essential, no excuses?
> 
> The LGP (or we) wasn't up to it but it was great learning experience for all 
> involved.  Good interpersonal story too about military/civilian world; and 
> about tiny-machine history c. 1959-81.
> 
> Techie about practical power of c. 1960 desk-size LGP-30 vs. mainframes. 
> IMHO LGP-30 and its ilk ushered in a new age, but it was hard to get 
> practical output.  However, if nobody had tried miniaturization and improved 
> software, we might not yet have personal computers.
> -0-
> 
> Story outline (which probably suffices but some other details are funnier 
> for anyone who either has or hasn't worked on or around an LGP-30):
> 
>    **  C. 1959 a Pentagon agency of the JCS acquired an LGP-30, and a USAF 
> tech sgt. who'd gone to LGP school to be the attendant.  The agency wanted 
> to make its LGP earn its keep, vs. paying big $X/hr. for IBM 704/709 time on 
> nearby thinktank's machines (under existing contract to C-E-I-R Inc., my 
> employer).
> 
>    **  Worldwide "nuclear targeting conference" needed ref. tables to assist 
> the targeteers, generally MAJ-COL ranks from all services who flew in for 
> the working conference to select next year's targets for US nukes and the 
> lat/long aimpoints.  Tables roughly involved probability of damage to 
> various target types as a function of weapon and aim-point distance 
> parameters.  The Pentagon office LGP-30 should be able to crank out that 
> pile of tables ready for duplication  --  right?
> 
>    **  I was still a rookie operations researcher and pseudo-math/statistics 
> guy,  But I was closest authority on the RAND equations that calculated such 
> nuclear damage estimates.  Handed the tables problem, within an intense few 
> days recast and invented math and flowcharted to compute specified output. 
> My C-E-I-R tech peers/srs. pronounced it sound after an intense blackboards 
> review.  So we knew we could oblige our Pentagon customer  --  at least 
> theoretically.
> 
>    **  The tech sgt. (let's call him Jack) and I were tasked to DO it, with 
> inflexible deadline of  0730 on a Mon. morning some 6-8 weeks hence.  We got 
> along fine  --  a good thing because over the final 2-5 weeks we spent more 
> time together than we did with our wives/kids.  He trusted my math and I 
> trusted his LGP programmer know-how, and we swapped ideas to work things 
> out.  (I never thru today was a programmer by trade, but somewhat by 
> osmosis.)
> 
>           Neither of us understood why LGP computation and paper tape I/O 
> were sooo slooow.  Just accepted that it was faster than monkeys at Friden 
> calculators/sliderules plus manual typewriters.
> 
>   **  Programming environment:  LGP's no-frills assembler (not even a macro 
> assembler, much less a compiler).  Only 16 primitive instructions so even I 
> could program the primitive math/logic math parts.  Supplemented by 
> floating-point subroutines  -- but IIRC no trig functions nor Exp/Ln either. 
> Writing/reading data to the drum (not RAM) was by word address, so 
> non-optimized for N words of data.  We two quickly identified those 
> bottlenecks.  E.g., I'd read about "spiraling" on disks and we adapted that 
> to "interlacing" data written to the drum.  We implemented non-arithmetic 
> math functions by exotic approximation formulas (ref. Cecil Hastings) or 
> table lookup+interpolation.
> 
>           Not enough time to do anything more clever algorithmically.  In 
> retrospect, I was thinking brute force of scientific mainframes and Jack was 
> just learning LGP-30's limitations
> 
>   **  We got all those foundations working in a couple weeks, and also the 
> flowcharted math equations.  Time to debug.  Jack debugged output format on 
> the slooow teletype as I debugged the math (comparing LGP output with my 
> hand-calculated test cases).
> 
>         Back in those stone ages, everyone (encouraged by IBM) believed that 
> testing was only to be sure.  No programmer nor computer nor sys software 
> could possibly make a mistake  -- right?  So Jack and I fell into the septic 
> tank of debugging.
> 
>    **  Our military superiors had walked in occasionally to see how our 
> project was going. Esp. the agency chief (a USMC COL) and his deputy (USA 
> LCOL).  They were content with progress and that Jack/I knew what we were 
> doing and exerting effort.  Even when we fell into testing septic tank, 
> everyone thought the breakthrough would come within hours.
> 
>    **  The breakthough came after a few all-nighters.  So we started the 
> LGP-30 chugging through all the combinations of parameters c. 0700 on a Sun. 
> morning.
> 
>           We'd make the Mon. 0730 deadline with time to spare.  Jack had 
> calculated that the production run would take 20 hrs. (assuming the LGP 
> didn't break down mechanically or blow a tube).
> 
>            Jack went off in Pentagon to find us celebratory coffee and 
> doughnuts.  I watched to be sure the teletype's paper didn't jam.  And idly 
> rechecked Jack's calculations for production run.
> 
>            Yikes!  An error by factor of 10.  Production run would take 200 
> hrs., not 20!!!
> 
>             Over coffee/doughnuts, Jack agreed.  He paled.  (BTW  --  Jack 
> was "Negro", a 1960 rarity in Pentagon much less as a USAF tech specialist.) 
> Since Jack feared court martial, I said I'd front a solution.  How about if 
> we do this?
> 
> **  My corp. superiors were unlikely to have authority without lots of phone 
> calls so c. 0830 I phoned the JCS agency deputy (John, LCOL USA) at home 
> number (which the agency duty officer gave me when I said it was an 
> emergency). To John, confessed a miscalculation over slooow LGP-30.
> 
>         Offered solution:  Call LGP sales and get list of all LGP-30s within 
> reasonable distance of Pentagon.  We can partition the parameters and send 
> program paper tapes to get 20-40 LGPs chugging in parallel to make the Mon. 
> deadline.  The calculations were classified Secret but there must be some 
> way to handle that and other logistics.
> 
>          John didn't laugh and he didn't rage.  We just discussed his idea 
> for a better solution.
> 
> **  Hats off to LCOL John and his behavior under fire at 0830 on Sun. 
> Clearly a cool military commander.
> 
>        (A year or so later I was named tech director and straw boss of the 
> large C-E-I-R project/contracts, age c. 27  --  a surprise to me and my 
> peers, not to mention to the PhDs who suddenly were nominally my 
> subordinates.  Confided to LCOL John that I didn't feel ready for all that. 
> John counseled me: "Bill,  just do what you've been doing and you'll be 
> fine. When I was 27 I received WW II battlefield promotion in Europe to 
> major.  Like you, didn't think I was ready but rose to the 
> responsibilities."  Ever since I've recalled John's advice when I feel 
> overwhelmed.)
> 
>      John at 0830 Sun. had a better solution than mine.  We discussed for 15 
> mins. and arrived at plan of action.  John's idea and command decision:
>          **  The targeting conference printed tables are crucial.  We wanted 
> to produce them on our in-house agency LGP-30, but forget that today.
>          **  I'll get you some skilled FORTRAN programmers.  You show them 
> what to calculate etc.  And you get your C-E-I-R mainframes fired up and get 
> us priority this Sun.
>          **  Can do?
> 
>     Can do, John, somehow or other with enough personal coffee and help from 
> C-E-I-R people, who often give your contracts priority.
> 
>       I asked: What about TSGT Jack, my teammate who has no mainframe 
> experience? John:  Take him along to C-E-I-R, Bill, if he'd be useful or 
> wants to watch.  Put him on phone and I'll tell him myself.  (Jack fearfully 
> talked to his LCOL.  His pallor subsided.  LCOL said Jack wasn't in trouble 
> and that his good work was learning experience that would make the LGP-30 
> more useful in future. Jack chose to spend Sun. at Baltimore home with 
> wife/kids, prolly napping after all-nighter.)
> -0-
> 
>   C. noon Sun. a USA major and 2 of his young noncoms (with master degrees 
> in math) met me at C-E-I-R.  We'd never met before but didn 't fool around 
> with formalities nor assessing ranks/credentials. Went directly to 
> blackboards where I explained the original objective and my 
> equations/flowchart.  They all knew math and FORTRAN (prolly better than I) 
> so that went fast.
> 
>  It took only 30-60 mins. of techie talk and Q&A.  Great guys who didn't act 
> as if they were in on Sun. to bail out snotty civilians. (Perhaps because 
> C-E-I-R culture wasn't snotty towards anyone seriously trying to solve a 
> problem.)
> 
> The MAJ and 2 noncoms started on FORTRAN coding pads.  I was avbl for 
> frequent ques. but otherwise I:  a)  Made sure an off-duty IBM 709 operator 
> had showed up and got the mainframe fired up for our exclusive use, on 
> demand;  b)  Keypunchers were ready to key FORTRAN coding pads into punched 
> cards;  c) Organized my test cases for the programmers; and d) Made sure 
> everyone including me was supplied with coffee/soft drinks and enough 
> blackboard chalk and coding pads.
> 
> By c. 3 pm the 3-man military programmer team had some FORTRAN code ready 
> for keypunch and trial compilations to fix syntax errors.  C-E-I-R 
> functionaries took care of that from coding sheets thru 709 output back to 
> the progammers' desks.   And then trying again with scribbled revisions. 
> I.e., an ad hoc team effort to serve the 3 most important guys in path to 
> success.  (C-E-I-R people were great too.)
> 
>  By c. 5 pm we had first successful FORTRAN compile and started on test 
> cases.
> 
> At 6:30 pm the whole apparatus recompiled and made its production run. It 
> took c. 4 mins. of IBM 709 time.  Output was magtape, which was printed on 
> off-line IBM 1401 to a high-speed lineprinter.
> 
> I forget how we got printout to Pentagon duplicators but it was in ample 
> time.  We cheered at c. 7 pm among ourselves but otherwise to Pentagon 
> superiors, just expected effort  --  now on to next tasks.
> 
> (BTW  --  IIRC no C-E-I-R nor Pentagon higher-up appeared and got in the 
> way.  IIRC they didn't phone either. Tacit assumption:  We were busy and 
> would call for help if we needed it.)
> 
> I.e., the LGP-30 effort took 6-8 weeks and failed because production run 
> would take 200 hrs.  The job was accomplished on a Sun. afternoon with 
> production run on IBM 709 of 3 mins. plus 1+ hr. of offline printing. 
> Mostly because of pure muscles of the 709;  and probably more important, its 
> programmer environment.
> 
>      (Recall that in 1960 FORTRAN still was horribly primitive, but it was 
> better than alternatives for ad hoc computations.  ALGOL was good but 
> implemented only on non-IBM computers.  COBOL was still in definition stage, 
> driven by pioneer Grace Hopper (CDR, USN).
> -0-
> 
> Never again touched that Pentagon LGP-30 but it taught me lots in 6-8 weeks. 
> Including how to make a machine do things when its instruction set is only 
> 16;  and its "RAM" is a mechanical magnetic drum;  and instruction execution 
> is measured in mutli-milleseconds; and I/O is via teletypewriter.
> 
>    (Learned that for ad hoc personal computation, hire muscle and skilled 
> programmers who use higher-level languages.  Bit-twiddling has its uses but 
> ...  often we just need to add  2+2 in fairly complex ways, and then add 
> 2.09 + Ln 5.73 tomorrow.
>    (In 1973 went thru intense design/computational project of 3+ mos;  and 
> though it was for big scientific computer company, couldn't afford the 
> delays to get computer answers. Calculated solo in a motel room with 
> sliderule, log tables and an early 4-function calculator.  To avoid that 
> ever again, spent much of my bonus on an early $650 HP-65 programmable 
> calculator with magnetic gumsticks.  And shortly when promoted to perform 
> those and other functions, I demanded a personal APL terminal  -- connected 
> to competitor APL service, not my coorpration's.  Got the terminal.  Ignored 
> big fuss raised by young UMass asst. prof. who was hired by my corp's data 
> services div. to spread APL gospel.  Used the APL terminal for 6-8 months, 
> by which time for misc. reasons corp. replaced it by personal PLATO 
> terminals for me, both in office and at home.  By 8-10 years later, personal 
> computers were $practical and software was becoming useful.
> -0-
> 
> See what the LGP-30 and a few other "desk" computers started?  IMHO they 
> deserve place in pantheon of computing history even if they didn't 
> accomplish practical work easily.
> 
> LGP-30 was ideological precursor IMHO to IBM 1620.  And then to the fine 
> minicomputers of DEC and its upstart competitor (both now RIP).
> -0-
> 
> People in that LGP-30 episode:
> 
>    **  Saw SSgt Jack in Pentagon occasionally.  We howdied warmly and asked 
> after  wives/kid.   Jack soon acquired another 1-2 stripes, at least one 
> IMHO due to LGP-30.
> 
>    **  Never again saw the MAJ and his noncoms IIRC.  Dunno even where they 
> came from on that Sun.  Hope we parted with as much mutual respect as I 
> think we deserved.  (Could be that military has math teams ready for ra;id 
> response, kinda like Spcial Forces.)
> 
>    **  LCOL John and his boss (USMC COL Ray) were very good guys and role 
> models.  A year+ later they recruited and interviewed me,  and offered me 
> job as tech dir. of what their agency had morphed into, the JCS real-time 
> Command Center's "support" center involving analyhsis+computing.  I 
> hesitated  before declining post that would by civil-service GS-15 
> equavalency made me a  COL before age 30.  Thanked Ray and John for their 
> trust  --  but I had family troubles, wasn 't sure I could play Pentagon 
> politics, and didn't want to become too committed to defense actitvies for 
> life.  They understood.  A few years later Ray recognized me at 
> Hartford-Springfield Airport and, after a few minutes catching up,  offered 
> my job at his retirement employer United Aircraft.   I said thanks but no 
> thanks.
>        Golly gosh  All very flattering from men I admired,
> -0-
> 
> C. 4,000 LGP-30s were manufactured starting 9/56, with price $40K.  In my 
> limited experience they were terrible for ad hoc computations  --  slow 
> hardware and weak software.
> 
> But I remember the LGP-30 fondly for all the other things it taught me. 
> Mainly how to overcome obstacles.
> 
> You, Ed, and others can draw your own morals.
> 
> ...  cheerio and smiles,  Bill
 

------------------------------------------------


There appear to be several villians in the tale
  a) Mis-sizing the computer (and its printing acapability)
     to the job at hand. If nothing else, I would suspect
     that the 10 char/second Frieden Flexowriter would 
     require service during a print load that would take a 
     1403 line printer "1+ hours".
     And it isn't quite fair to compair a 709 "super computer"
     with hardware floating point with the minimal LGP-30.
     The poor little guy never had a chance with a big league challenge ;-))
  b) in 1959, the FORTRAN like ACT III was available for the LGP 30.
     The powers that were should have acquired it 
     for a few bucks rather than use expensive civilian labor
     to hand code some $50,000 machine.

     I used ACT III on the college LGP-30 in 1959, kinda fun.
     It was a definite improvement over the original
     FORTRAN  !!  Nicer formatting, automatic mixed mode, ...

   c) I'm am presuming that there was considerable
     floating point required in your computations.
     If you had to write your own, your code was not likely 
     optimized for the drum rotation.  As you know, the LGP-30
     had a cute 7 to 1 address  scheme to help some.

     The floating point package that came with your LGP-30,
     hopefully, was somewhat adapted for rotational latency.

   d) I guess that was a "character building" experience,
         like "boot camp"?   ;-))

Glad you came out smelling like a rose,
  even if magic was required.
Organizations are often defensive critters, ready to circle the 
waggons.  All in the organization rise to help show that the 
  organization is not flawed.
Works from the local library organization all the way to the the U.N.
  (Say the oil-for-food program.)


---------------------

My first post-college interaction with LGP-30s was
to install a GE-225 (core machine with 21 us memory cycle time)
and an attached AAU with floating point capability at AirProducts.
They were using a pair of LGP-30s to do heat-exchanger 
design, and the poor little fixed point drum machines were
no longer adequate.

To sell Air Products the CE225, an LGP-30 emulator was
promised and delivered. Along with paper tape I/O, not
to common on GE-225s.  The emulator worked perfectly
right out of the box, and when adapted to shunt the
LGP-30 floating point calls to the AAU - what took
an LGP-30 all night to do was done in like 10 seconds
on the core machine with the hardware floating point.

Those poor little LGP-30s were gone in a week.
I felt kinda sad the little guys didn't even have 
a going away party :-((     ;-))

--------------------------------------

Ed Thelen